fix(scenario): coord-cover-and-move — recalibrate after engine movement fixes

openra_bench/scenarios/packs/coord-cover-and-move.yaml

@@ -14,18 +14,20 @@
 # also cross. Both squads converge on the eastern gathering region
 # having routed AROUND the fire zone rather than through it.
 #
-# Engine balance pass (armor-class weapon selection + stance + parallel
-# production + pbox-fires fixes): the original design assumed a dense
-# anti-tank-infantry cluster could out-trade a 6-tank charge. Post-fix,
-# medium tanks decisively OUT-DPS infantry — a 6-tank column charging
-# even a 24-strong e3 swarm one-shots the rockets and crosses with
-# ZERO losses, collapsing the "mass-charge loses" discriminator. The
-# fire zone is therefore now ANCHORED BY 4tnk HEAVY TANKS (1 on easy,
-# 2 on medium/hard): the heavy-tank brawl is what genuinely punishes a
-# centre charge, while the e3/e1 infantry keep the anti-tank-infantry
-# doctrine and the range geometry meaningful. All enemy units are
-# stance:2 Defend (auto-fire in range but STATIONARY) so a staller is
-# a clean timeout LOSS rather than chased down by hunters.
+# Engine movement-fix recalibration ((A) attack_unit on out-of-sight
+# targets paths normally — no teleport; (B) a moving unit fires AND
+# takes fire en route — no sprint-invincibility): a 6-tank column that
+# attack_moves straight through the centre now genuinely trades fire
+# with the cluster while crossing. The load-bearing punisher is the
+# combination of e3 ANTI-TANK ROCKET soldiers (the real anti-armour
+# DPS) AND 4tnk HEAVY-TANK anchors: probing shows e3 rockets alone or
+# a single 4tnk only cost a charging column ONE tank, but FOUR e3 plus
+# TWO 4tnk reliably cost it TWO. Every level therefore fields an e3
+# rocket cluster anchored by TWO 4tnk (easy uses a smaller 3× e3
+# cluster; medium/hard a 4× e3 cluster, hard adds e1 rifle screens).
+# All enemy units are stance:2 Defend (auto-fire in range but
+# STATIONARY) so a staller is a clean timeout LOSS rather than being
+# chased down by hunters.
 #
 # Real-world anchor: US Army FM 3-21.8 bounding overwatch — one fire
 # team in a stationary OVERWATCH posture provides suppressive fire
@@ -39,8 +41,8 @@
 #   • stall (observe-only): no squad crosses → win region empty →
 #     after_ticks timeout LOSS.
 #   • both-charge-together: all 6 tanks rush a single eastward column
-#     straight through the centre fire zone; the heavy-tank brawl plus
-#     the rocket fire bust the attrition cap → LOSS.
+#     straight through the centre fire zone; the rocket fire plus the
+#     heavy-tank brawl bust the attrition cap → LOSS.
 #   • single-squad-only: send Squad A through alone (Squad B idle
 #     west); A absorbs the centre fire AND half the force never
 #     reaches the win region (n=4 region clause unreachable) → LOSS.
@@ -52,16 +54,17 @@
 #     deadline losing ≤ the cap → WIN.
 #
 # Why the spec works (engine combat sheet):
-#   • 4tnk = heavy/mammoth tank — high HP and a heavy cannon; its
-#     brawl is what kills ≥2 of a 6-tank column that charges through
-#     the centre. Held at the cluster centre (y=20) it reaches a
-#     charging column on the y≈14..16 / y≈24..26 lanes.
-#   • e3 = anti-tank rocket soldier (rng4c, sight5c) — reinforces the
-#     centre fire zone and keeps the overwatch range geometry (a tank
-#     marginally out-ranges e3) meaningful, but does NOT on its own
-#     out-trade a tank column.
+#   • e3 = anti-tank rocket soldier (rng4c, sight5c) — the real
+#     anti-armour DPS of the fire zone; a 4× e3 cluster plus the two
+#     heavy-tank anchors is what kills ≥2 of a 6-tank column charging
+#     through the centre.
+#   • 4tnk = heavy/mammoth tank — high HP and a heavy cannon; the
+#     SECOND 4tnk anchor is what pushes a charging column's losses
+#     from one tank to two. Held at the cluster centre (y=20) it
+#     reaches a charging column on the y≈14..16 / y≈24..26 lanes.
 #   • 2tnk = medium tank (rng4.75c, sight6c, hp400) — out-DPSes
-#     infantry but is brawled down by the 4tnk anchor in the centre.
+#     infantry but is ground down by the e3 rockets + 4tnk anchors
+#     when it crosses the centre.
 #   • Periphery geometry: the centre fire zone occupies y≈13..27. A
 #     squad that rises to y≈6 (NORTH) or y≈34 (SOUTH) BEFORE reaching
 #     the fire zone's longitude (x≈50) routes entirely outside the
@@ -133,30 +136,33 @@ base:
 
 levels:
   # ── EASY ─────────────────────────────────────────────────────────
-  # Bare bound-and-cover skill: a 4× e1 rifle-infantry cluster at the
-  # centre anchored by ONE 4tnk heavy tank. Engine balance pass: the
-  # post-armor-fix infantry can no longer kill a column of 6 medium
-  # tanks (they out-DPS the e1/e3 cluster outright), so the fire zone
-  # is anchored by a 4tnk heavy tank — its brawl is what punishes a
-  # mass charge. A both-charge straight through the centre clips the
-  # 4tnk's brawl and loses ≥2 tanks; stall times out; single-squad
-  # leaves half the force idle. The intended bound-and-cover routes
-  # both squads through the far periphery (y≈6 / y≈34, outside the
-  # centre fire zone) and converges east — taking zero losses.
+  # Bare bound-and-cover skill: a 3× e3 anti-tank-rocket cluster at the
+  # centre anchored by TWO 4tnk heavy tanks. Engine movement-fix
+  # recalibration: a charging column now genuinely trades fire while
+  # crossing — probing shows e3 rockets are the real anti-armour DPS,
+  # but it takes TWO heavy-tank anchors alongside them to cost a
+  # 6-tank charge TWO tanks. A both-charge straight through the centre
+  # eats the e3 rockets + the heavy-tank brawl and loses 2 tanks
+  # (busts the cap); stall times out; single-squad leaves half the
+  # force idle AND loses 2 crossing. The intended bound-and-cover
+  # routes both squads through the far periphery (y≈6 / y≈34, outside
+  # the centre fire zone) and converges east — taking zero losses.
+  # Easy is the FORGIVING tier: a smaller 3× e3 cluster (medium/hard
+  # field 4× e3), so the bounding-lane geometry is the bare skill.
   easy:
     description: >
       Two squads (Squad A at (10,15), Squad B at (10,25), 3 medium
-      tanks each) must cross a centre-of-map fire zone — a rifle-
-      infantry (e1) cluster anchored by one heavy tank (4tnk) at
-      roughly (50,20) — then land ≥4 tanks in the EAST gathering
-      region (x ≈ 100, y ≈ 20, radius 6) before tick 4500, losing ≤1
-      tank in total. Charging both squads straight through the centre
-      drives them into the heavy tank's brawl and busts the cap; a
-      single column alone absorbs every shot AND leaves half the
-      force behind. Alternate ROLES: one squad anchors at the edge of
-      enemy range to draw fire while the other sweeps through the
-      periphery (y outside the centre fire zone) and crosses; then
-      switch.
+      tanks each) must cross a centre-of-map fire zone — a small
+      anti-tank rocket-soldier (e3) cluster anchored by two heavy
+      tanks (4tnk) at roughly (50,20) — then land ≥4 tanks in the
+      EAST gathering region (x ≈ 100, y ≈ 20, radius 6) before tick
+      4500, losing ≤1 tank in total. Charging both squads straight
+      through the centre drives them into the rocket fire and the
+      heavy-tank brawl and busts the cap; a single column alone
+      absorbs every shot AND leaves half the force behind. Alternate
+      ROLES: one squad anchors at the edge of enemy range to draw
+      fire while the other sweeps through the periphery (y outside
+      the centre fire zone) and crosses; then switch.
     overrides:
       actors:
         # Squad A — northern fire team, staged at (10, 15).
@@ -167,17 +173,17 @@ levels:
         - {type: 2tnk, owner: agent, position: [10, 24], stance: 1}
         - {type: 2tnk, owner: agent, position: [10, 25], stance: 1}
         - {type: 2tnk, owner: agent, position: [10, 26], stance: 1}
-        # Centre fire zone: 4× rifle infantry + 1 heavy-tank anchor,
-        # all stance:2 Defend (auto-fire in range but STATIONARY — so
-        # a stall policy is a clean timeout LOSS, not an attrition
-        # one). Cells avoid (50,20) — CLAUDE.md footgun: enemy
-        # clusters at (50,20) silently fail to place. Off-axis
+        # Centre fire zone: 3× anti-tank rocket soldiers + 2 heavy-
+        # tank anchors, all stance:2 Defend (auto-fire in range but
+        # STATIONARY — so a stall policy is a clean timeout LOSS, not
+        # an attrition one). Cells avoid (50,20) — CLAUDE.md footgun:
+        # enemy clusters at (50,20) silently fail to place. Off-axis
         # neighbours (49/51 × 19/21) are verified-placeable.
-        - {type: e1,   owner: enemy, position: [49, 19], stance: 2}
-        - {type: e1,   owner: enemy, position: [49, 21], stance: 2}
-        - {type: e1,   owner: enemy, position: [51, 19], stance: 2}
-        - {type: e1,   owner: enemy, position: [51, 21], stance: 2}
-        - {type: 4tnk, owner: enemy, position: [50, 20], stance: 2}
+        - {type: e3,   owner: enemy, position: [49, 19], stance: 2}
+        - {type: e3,   owner: enemy, position: [49, 21], stance: 2}
+        - {type: e3,   owner: enemy, position: [51, 21], stance: 2}
+        - {type: 4tnk, owner: enemy, position: [49, 20], stance: 2}
+        - {type: 4tnk, owner: enemy, position: [51, 19], stance: 2}
         # Persistent far enemy `fact` — keeps the episode alive past
         # the central cluster being cleaned out so the win/fail
         # evaluator sees the terminal frame (no DRAW collapse).
@@ -195,14 +201,15 @@ levels:
     max_turns: 50
 
   # ── MEDIUM ───────────────────────────────────────────────────────
-  # +1 axis vs easy: ANTI-TANK rocket soldiers (e3) replace the rifle
-  # infantry AND the heavy-tank anchor doubles to TWO 4tnk. The fire
-  # zone is genuinely lethal to a 2tnk column crossing the centre.
-  # Both-charge-together LOSES (the centre brawl busts units_lost_
-  # lte:1); single-squad alone LOSES (lone column absorbs the centre
-  # fire AND half the force never reaches the win region). Only true
-  # bound-and-cover (route the squads through the far periphery,
-  # alternating cover/move roles) lands ≥4 tanks losing ≤1.
+  # +1 axis vs easy: the e3 anti-tank rocket cluster grows from 3 to
+  # 4 (a wider, denser fire zone) while keeping the TWO 4tnk anchors.
+  # The fire zone is genuinely lethal to a 2tnk column crossing the
+  # centre. Both-charge-together LOSES (the centre brawl + rockets
+  # bust units_lost_lte:1); single-squad alone LOSES (lone column
+  # absorbs the centre fire AND half the force never reaches the win
+  # region). Only true bound-and-cover (route the squads through the
+  # far periphery, alternating cover/move roles) lands ≥4 tanks
+  # losing ≤1.
   medium:
     description: >
       Two squads (Squad A at (10,15), Squad B at (10,25), 3 medium

tests/test_coord_cover_and_move.py

@@ -3,7 +3,7 @@
 The bar (binding):
 - stall LOSES on every level + every seed (after_ticks timeout);
 - both-charge-through-the-centre LOSES on every level + every seed
-  (the heavy-tank brawl busts the attrition cap);
+  (the rocket fire + heavy-tank brawl bust the attrition cap);
 - single-squad-only LOSES on every level + every seed (half the force
   idle → the n=4 region clause is unreachable);
 - the intended bound-and-cover (periphery route around the fire zone)
@@ -12,18 +12,20 @@ The bar (binding):
   clause (no DRAW degeneracy);
 - hard ships ≥2 `spawn_point` groups (seed-driven start variation).
 
-Recalibrated after the engine balance pass (armor-class weapon
-selection + stance + parallel-production + pbox-fires fixes): the
-original design assumed a dense anti-tank-infantry cluster could
-out-trade a 6-tank charge, but post-fix medium tanks decisively
-out-DPS infantry — a 6-tank column charging even a 24-strong e3
-swarm one-shots the rockets and crosses with ZERO losses, collapsing
-the "mass-charge loses" discriminator. The fire zone is now ANCHORED
-BY 4tnk HEAVY TANKS (1 easy / 2 medium+hard): the heavy-tank brawl is
-what punishes a centre charge, while the e3/e1 infantry keep the
-anti-tank-infantry doctrine and the range geometry meaningful. All
-enemy units are stance:2 Defend (auto-fire in range, STATIONARY) so a
-staller is a clean timeout LOSS, not chased down by hunters.
+Recalibrated after the engine movement fixes ((A) attack_unit on
+out-of-sight targets paths normally — no teleport; (B) a moving unit
+fires AND takes fire en route — no sprint-invincibility). A 6-tank
+column that attack_moves through the centre now genuinely trades fire
+while crossing. Probing showed the e1 rifle cluster on the old easy
+tier could no longer punish the charge at all (0 losses): the
+load-bearing punisher is e3 ANTI-TANK ROCKET soldiers (the real
+anti-armour DPS) PLUS two 4tnk HEAVY-TANK anchors — e3 rockets alone,
+or a single 4tnk, only cost a charge ONE tank; four e3 plus two 4tnk
+cost it TWO. Every level now fields an e3 rocket cluster anchored by
+TWO 4tnk (easy a smaller 3× e3 cluster, medium/hard a 4× e3 cluster,
+hard adds e1 rifle screens). All enemy units are stance:2 Defend
+(auto-fire in range, STATIONARY) so a staller is a clean timeout
+LOSS, not chased down by hunters.
 
 Validation is scripted (no model / network).
 """
@@ -187,36 +189,39 @@ def test_two_squads_six_tanks_each_level():
     )
 
 
-def test_fire_zone_uses_anti_tank_rocket_on_medium_and_hard():
-    """The fire zone's lethality is the load-bearing property —
-    medium / hard must use e3 (anti-tank rocket soldier) at the
-    centre cluster. Easy may use e1 (forgiving)."""
+def test_fire_zone_uses_anti_tank_rocket_every_level():
+    """The fire zone's lethality is the load-bearing property — post
+    engine movement fixes the e3 (anti-tank rocket soldier) cluster is
+    the real anti-armour DPS, so EVERY level uses e3 at the centre.
+    Easy fields a smaller cluster (≥3 e3, the forgiving tier);
+    medium / hard a denser one (≥4 e3)."""
     pack = load_pack(PACK_PATH)
-    for lvl in ("medium", "hard"):
+    for lvl, want_e3 in (("easy", 3), ("medium", 4), ("hard", 4)):
         c = compile_level(pack, lvl)
         types = [a.type for a in c.scenario.actors if a.owner == "enemy"]
-        assert types.count("e3") >= 4, (
-            f"{lvl}: need ≥4 e3 (anti-tank rocket) for fire zone; got {types}"
+        assert types.count("e3") >= want_e3, (
+            f"{lvl}: need ≥{want_e3} e3 (anti-tank rocket) for fire "
+            f"zone; got {types}"
         )
         # Persistent far enemy marker (engine auto-done mitigation).
         assert "fact" in types, f"{lvl}: needs persistent enemy fact"
 
 
-def test_fire_zone_anchored_by_heavy_tank():
-    """Post-engine-balance-pass invariant: medium tanks out-DPS
-    infantry, so the centre fire zone is anchored by 4tnk heavy
-    tank(s) — the heavy-tank brawl is what punishes a centre charge.
-    Every level must carry at least one enemy 4tnk; medium and hard
-    carry two."""
+def test_fire_zone_anchored_by_two_heavy_tanks():
+    """Recalibration invariant: probing showed the e3 rockets alone or
+    a single 4tnk only cost a 6-tank charge ONE tank — it takes TWO
+    4tnk anchors alongside the rocket cluster to reliably cost the
+    charge TWO. Every level therefore anchors the fire zone with two
+    enemy 4tnk heavy tanks."""
     pack = load_pack(PACK_PATH)
-    for lvl, want in (("easy", 1), ("medium", 2), ("hard", 2)):
+    for lvl in ("easy", "medium", "hard"):
         c = compile_level(pack, lvl)
         n4tnk = sum(
             1 for a in c.scenario.actors
             if a.owner == "enemy" and a.type == "4tnk"
         )
-        assert n4tnk >= want, (
-            f"{lvl}: fire zone must be anchored by ≥{want} enemy 4tnk "
+        assert n4tnk >= 2, (
+            f"{lvl}: fire zone must be anchored by ≥2 enemy 4tnk "
             f"(heavy-tank brawl), got {n4tnk}"
         )